clinical study comparative diagnostic accuracy of ganglion ...clinical study comparative diagnostic...

Clinical StudyComparative Diagnostic Accuracy of Ganglion Cell-InnerPlexiform and Retinal Nerve Fiber Layer Thickness Measures byCirrus and Spectralis Optical Coherence Tomography inRelapsing-Remitting Multiple Sclerosis

Julio J. González-López,1,2,3 Gema Rebolleda,1,2 Marina Leal,1 Noelia Oblanca,1

Francisco J. Muñoz-Negrete,1,2 Lucienne Costa-Frossard,4 and José C. Álvarez-Cermeño4

1 Department of Ophthalmology, Ramon y Cajal University Hospital, Carretera de Colmenar Km 9, 1, 28034 Madrid, Spain2Department of Surgery, Alcala de Henares University, Madrid, Spain3Medical Retina Department, Moorfields Eye Hospital, London, UK4Department of Neurology, Ramon y Cajal University Hospital, Madrid, Spain

Correspondence should be addressed to Gema Rebolleda; [email protected]

Received 29 July 2014; Revised 23 August 2014; Accepted 26 August 2014; Published 14 September 2014

Academic Editor: Antonio Ferreras

Copyright © 2014 Julio J. Gonzalez-Lopez et al.This is an open access article distributed under the Creative Commons AttributionLicense, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properlycited.

Objective. To estimate sensitivity and specificity of several optical coherence tomography (OCT) measurements for detectingretinal thickness changes in patients with relapsing-remitting multiple sclerosis (RRMS), such as macular ganglion cell-innerplexiform layer (GCIPL) thickness measured with Cirrus (OCT) and peripapillary retinal nerve fiber layer (pRNFL) thicknessmeasured with Cirrus and Spectralis OCT.Methods. Seventy patients (140 eyes) with RRMS and seventy matched healthy subjectsunderwent pRNFL and GCIPL thickness analysis using Cirrus OCT and pRNFL using Spectralis OCT. A prospective, cross-sectional evaluation of sensitivities and specificities was performed using latent class analysis due to the absence of a goldstandard. Results. GCIPL measures had higher sensitivity and specificity than temporal pRNFL measures obtained with bothOCT devices. Average GCIPL thickness was significantly more sensitive than temporal pRNFL by Cirrus (96.34% versus 58.41%)and minimum GCIPL thickness was significantly more sensitive than temporal pRNFL by Spectralis (96.41% versus 69.69%).Generalised estimating equation analysis revealed that age (𝑃 = 0.030), optic neuritis antecedent (𝑃 = 0.001), and diseaseduration (𝑃 = 0.002) were significantly associated with abnormal results in average GCIPL thickness. Conclusion. Average andminimum GCIPL measurements had significantly better sensitivity to detect retinal thickness changes in RRMS than temporalpRNFL thickness measured by Cirrus and Spectralis OCT, respectively.

1. Introduction

Relapsing-remitting multiple sclerosis (RRMS) is a chronic,immune-mediated demyelinating disease of the central ner-vous system that frequently involves the visual pathways,usually in the form of optic neuritis (ON) [1]. Postmortemanalysis revealed optic nerve lesions in 94–99% of RRMSpatients, even in the absence of a clinical history of ON [2].

Optical coherence tomography (OCT) is a noninvasiveand reproducible tool for evaluating the retinal and opticdisc anatomy of patients with this clinical disorder. It uses

low-coherence interferometry to obtain detailed images ofthe retinal architecture. Modern high-speed spectral-domain(SD) OCT devices can obtain high resolution images ofthe retina. Computerised algorithms can be used on theseimages in order to automatically identify and obtain thicknessmeasurements of discrete retinal layers, including the retinalnerve fiber layer (RNFL) and the macular ganglion cell-innerplexiform layers (GCIPL) [3].

In patients with RRMS, the main focus has been the eval-uation of the peripapillary retinal nerve fiber layer (pRNFL).The RNFL contains the unmyelinated axons originating from

Hindawi Publishing CorporationBioMed Research InternationalVolume 2014, Article ID 128517, 10 pageshttp://dx.doi.org/10.1155/2014/128517

2 BioMed Research International

the ganglion cell neurons. In a previous report, we foundthat Spectralis showed a significantly higher thinning fortemporal quadrant than Cirrus in eyes of RRMS patients,suggesting that N-site axonal analysis could define axonaldamage in relapsing-remitting multiple sclerosis patientsearlier than conventional pRNFL analysis [4].

Optic nerve demyelination, due to clinical or subclinicalON, can result in retrograde degeneration of the optic nerveaxons, leading to RNFL and GCIPL thinning [5, 6]. In fact,several studies have reported statistically significant thinningof the pRNFL and GCIPL in patients with RRMS with andwithout optic neuritis compared to healthy control subjects[7–10].

Moreover, macular GCIPL thickness has been foundto have better structure-function correlations than pRNFLthickness with both visual function and disability in RRMSpatients [11]. At least in part, this observation may be due tothe superior reproducibility of GCIPL over pRNFL thicknessmeasurements [11].

Thus, we hypothesize that GCIPL measurements canbe more sensitive to and specific of retinal involvement inpatients with RRMS than those with pRNFL measurements.

The main purpose of this study was to estimate thesensitivity and specificity of macular GCIPL thickness mea-sured with the Cirrus OCT ganglion cell analysis (GCA)algorithm (Carl Zeiss Meditec AG, Jena, Germany) andpRNFL thickness analysis with Cirrus and Spectralis (Hei-delberg Engineering GmbH, Heidelberg, Germany) OCTsin detecting retinal thickness changes in eyes from patientswith a clinical diagnosis of RRMS versus age-matched normalsubjects using latent class analysis.

Ancillary objectives were to quantify color-code abnor-malities in GCIPL and pRNFL measures by Cirrus andSpectralis OCTs in eyes frompatients with RRMS and healthycontrol subjects, to quantify the correlations of GCIPL andpRNFLmeasurements with the visual function and disabilityin MS patients, and to study the effect of optic neuritisantecedent on the obtained measurements.

2. Methods

2.1. Subjects. An observational, prospectively recruited,cross-sectional studywas performed.The studywas approvedby the Research Ethics Committee in the Ramon y Cajal Uni-versity Hospital. All research complied with the tenets of thedeclaration of Helsinki, and all subjects participating in thestudy gave their written informed consent. Confidentialityof participating subjects was protected throughout the study.

We included 70 patients with a diagnosis of RRMS and 70healthy control subjects. Patients were enrolled consecutivelyfrom the Neuroophthalmology Department from January2012 to September 2012. Healthy controls without a historyof neurological and ophthalmological disease were recruitedamong the hospital staff.

Diagnosis of RRMS was based on McDonald criteria bythe treating neurologist [12]. None of the included patientshad a diagnosis of secondary progressive multiple sclerosis.

All participants underwent a complete neuroophthalmicevaluation that included pupillary, anterior segment, andfunduscopic examinations; assessment of logMAR best cor-rected visual acuity (BCVA), and they were scanned afterpupillary dilation with Cirrus (Carl Zeiss Meditec AG, Jena,Germany) and Spectralis (Heidelberg Engineering GmbH,Heidelberg, Germany) OCTs on the same day in randomorder. Both eyes of each subject were included. Exclusioncriteria were intraocular pressure of 21mmHg or higher, anoptic disc suspicious for glaucoma under dilated funduscopy,a refractive error greater than 5.0 diopters (D) of sphericalequivalent or 3.0D of astigmatism in either eye, mediaopacity, a recent history of optic neuritis 6 months prior tothe day of imaging, systemic conditions that could affect thevisual system, a history of ocular trauma, or concomitantocular diseases, including glaucoma.

Related medical records were carefully reviewed, includ-ing the duration of the disease, the ExpandedDisability StatusScale (EDSS) scored by a neurologist (LC), and the presenceof prior episodes of optic neuritis (ON) as reported by thetreating neurologist and the patient.

The visual field (VF)was tested only in the eyes of patientswith RRMS, using a Humphrey Field Analyzer (Carl ZeissMeditec AG, Jena, Germany) and the SITA Standard protocol(program 24-2). VF test was considered reliable when fixationlosseswere less than 20%and false-positive and false-negativeerrors were less than 15%.

2.2. Optical Coherence Tomography Measurements. A single,well-trained optometrist (NO) performed all OCT exami-nations in random order to prevent any fatigue bias. Allpoor-quality scans were rejected, defined as those with signalstrength of ≤6 by Cirrus. For Spectralis OCT only thoseimages with a signal-to-noise score higher than 25 dB wereanalyzed. Scans with misalignment, segmentation failure,decentration of the measurement circle, and poor illumina-tion or those out of focus were excluded from the analysis.Thus, manual correction of plotting errors of automatedsegmentation was not performed in this study.

Methodology for pRNFL imaging inCirrus and Spectralishas been reported previously [3]. Briefly, cross-sectionalimaging of the peripapillary area was performed using CirrusOCT. pRNFL thickness was determined using the optic disccube protocol (software version 5.1.1.6) that generates a cubeof data through a 6mm square grid. A 3.46mm in diametercalculation circle was automatically positioned around theoptic disc. Cirrus OCT provides average pRNFL thicknessand maps with 4 quadrants (superior, inferior, nasal, andtemporal) and 12 clock hours, including classification froman internal normative database.

Spectralis OCT (software version 5.2.0.3) simultaneouslycaptures infrared fundus and SD-OCT images at 40,000 A-scans per second. A real-time eye-tracking system measureseye movements and provides feedback to the scanningmechanism to stabilize the retinal position of the B-scan.The instrument uses 1024 A-scan points from a 3.45mmcircle centered on the optic disc. The examiner is requiredto manually place the scan around the optic disc. Peripap-illary RNFL measurements were obtained using the N-site

BioMed Research International 3

(a)

80

78

78

6967

72

(b)

225

150

75

0

(𝜇m

)

(c)

(d)

Figure 1: Ganglion cell-inner plexiform layer (GCIPL) analysis of the left eye of a patient with relapsing-remitting multiple sclerosis withoutoptic neuritis antecedent in a 6 × 6 × 2mm macular cube using a Cirrus optical coherence tomography. (a) Deviation map of the GCIPLthickness (red: below percentile 1; yellow: below percentile 5). (b) Sector distribution. (c) GCIPL thickness map, with overlying ellipsesshowing the dimensions of the analyzed annulus.Theouter ellipse has a vertical diameter of 4mmand the inner ellipse of 1mm. (d)HorizontalB-scan centered in the fovea, showing the automated segmentation of the GCIPL.

axonal protocol, which differs from the standard pRNFL scanbecause it starts and terminates in the nasal side of opticnerve. Scans were obtained using the high resolution (HR)mode and using automatic real-time (ART) for averaging9 B-scan frames in order to improve image quality. ThepRNFL Spectralis protocol generates a map showing theaverage thickness, maps with 4 quadrants (superior, inferior,nasal, and temporal), and maps with 6 sector thicknesses(superonasal, nasal, inferonasal, inferotemporal, temporal,and superotemporal).

The pRNFL thicknesses in the normal range are repre-sented by green backgrounds. Those that are abnormal at the5% and at the 1% level are represented by yellow and redbackgrounds, respectively. The hypernormal (95th to 100thpercentiles) pRNFL thicknesses are presented by awhite colorin Cirrus and by a blue/purple color in Spectralis.

Cross-sectional imaging of the macular area was per-formed using Cirrus OCT macular cube (512 × 128). Thisacquisition protocol generates a cube through a 6mm squaregrid of 128 B-scans of 512 A-scans each. A built-in GCIPLanalysis algorithm detects and measures the thickness of themacular GCIPL within a 6 × 6 × 2mm elliptical annulusarea centered on the fovea. The annulus has an inner verticaldiameter of 1mm, which was chosen to exclude the portionsof the fovea where the layers are very thin and difficult todetect accurately, and an outer vertical diameter of 4mm,whichwas chosen according towhere theGCL again becomes

thin and difficult to detect. The GCA algorithm identifiesthe outer boundaries of the RNFL and IPL. The differencebetween the RNFL and the IPL outer boundary segmenta-tions yields the combined thickness of the RGC layer and IPL.Cirrus OCT provides quantitative assessment of the ganglioncell and inner plexiform layers (GCIPL) in 6 circular sectorscentered in the fovea (superonasal, superior, inferonasal,inferotemporal, inferior, and superotemporal). It also givesinformation on the mean and minimum GCIPL thicknessfor each eye and compares these figures with a normativedatabase (Figure 1). The GCIPL thicknesses in the normalrange are represented by green backgrounds. Those that areabnormal at the 5% and at the 1% level are represented byyellow and red backgrounds, respectively. The hypernormal(95th to 100th percentiles) pRNFL thicknesses are presentedby a white color.

2.3. Statistical Analysis. Data were analyzed using Stata/SE12.0 for Unix and IBM SPSS Version 20 for Unix. A 𝑃 value ofless than 0.05 was considered statistically significant.

Quantitative variables were summarized as mean ± stan-dard deviation. Qualitative variables were summarized asabsolute value (percentage). Generalized estimating equationmodels accounting for sex, age, and within-patient intereyecorrelations were used to examine correlations and associa-tions between variables.


When evaluating new medical diagnostic tests, data maybe obtained from one or more tests, but none of these can beconsidered a gold standard, that is, a diagnostic test with 100%sensitivity and specificity [13]. Latent class analysis (LCA)is based on the concept that observed results of differentimperfect tests for the same disease are influenced by a latentcommon variable, the true disease status, which cannot bedirectly measured. In a group of patients with unknowndisease status, for whom results from several diagnostictests are available, LCA will model the probability of eachcombination of test results on the latent class and willprovide an estimate of sensitivity and specificity for each ofthe diagnostic tests evaluated [14, 15]. LCA has been usedextensively for the estimation of sensitivity and specificityof diagnostic tests in the absence of a valid gold standard,mainly in microbiology [16, 17] and psychology [18], but alsoin ophthalmology [19].

In this study, we implemented the basic latent classmodel,using the assumption of conditional independence giventhe latent class. In basic LCA, there are no associationsbetween the observed variables within each category of thelatent variable. The latent variable is the true status onthe disease, and the hypothesis is that there are two latentclasses (presence or absence of retinal thickness changes).As more than one pRNFL measure could not be fittedinto the same model due to the conditional independenceassumption, two LCAmodels were built. Four variables wereincluded in each LCA; average andminimummacularGCIPLthicknesses by Cirrus OCT and BCVA were present in bothmodels; temporal pRNFL thicknesses by Cirrus OCT orby Spectralis OCT were present in one model each. LCArequires tests with binary outcomes to create the model. Forsimplification of the analysis, white, blue, purple, and greensectors have been labeled as “normal,” and yellow and redones as “abnormal.” For BCVA, values better than or equalto 0.3 LogMAR were labeled as normal and those worse than0.3 as “abnormal.” BCVA was included in the model in orderto provide a functional outcome that could help better definethe latent class “retinal thickness change.” Temporal pRNFLwas selected as it was the quadrant with a higher frequencyof pRNFL thinning and abnormal results in previous studies[4, 20–23].

LCA was performed using TAGS software implementedin R version 2.2 (R Development Core Team and R Foun-dation for Statistical Computing, 2005). The fit of LCAmodel for the assumption of conditional independence wasperformed through the goodness-of-fit test followed by theevaluation of residual correlations between tests.

3. Results

SeventyRRMSpatients and seventy age- and gender-matchedhealthy controls were enrolled in the study. All participantswere of Caucasian descent. Table 1 summarizes the demo-graphic and clinical characteristics of the participants.

Overall, average pRNFL and temporal quadrant pRNFLthickness by both Cirrus and Spectralis OCTs were signifi-cantly lower in both ON and non-ON RRMS eyes compared

to healthy eyes (𝑃 < 0.001). Similarly, average,minimum, andeach of the 6 sectors GCIPL thicknesses yielded by Cirruswere significantly lower in RRMS compared to healthy eyesin both ON and non-ON eyes (𝑃 < 0.001).

All these measurements were significantly lower in eyeswith a prior history of ON compared to non-ON eyes (𝑃 <0.001).

Table 2 shows the percentage of abnormal color-codedmeasurements (defined as red or yellow color codes) obtainedbyGCIPL and pRNFL analysis in healthy andRRMS patients.Abnormal results were significantly more common in ONand non-ON RRMS eyes versus healthy eyes and in eyes withONantecedent versus thosewithout this antecedent in RRMSpatients.

Overall, the highest abnormal percentage was observedin minimum (47.8%) followed by average (46.4%) GCIPLanalysis. The sector in GCIPL test showing the highestabnormality rate was the superonasal (47.1%) followed bysuperotemporal sector (45.7%). The abnormality rates weresignificantly higher in eyes with a priorON compared to non-ON eyes (Figure 2).

Using Cirrus OCT, average GCIPL was altered morefrequently than average pRNFL in eyes of patientswithRRMS(46% versus 33%, resp.; 𝑃 < 0.001).

In a subgroup analysis comparing abnormal resultsbetween GCIPL and pRNFL by ON antecedent, averageand minimum GCIPL measurements yielded the highestabnormal results for both ON and non-ON eyes (Table 2).

Table 3 shows the estimated sensitivity and specificity todetect retinal thickness changes by OCT with the two LCAmodels. The test for evaluating the fit of the model withconditional independence (goodness-of-fit test) proved to beadjusted (P value = 0.938 for model A and 0.836 for modelB). The residual correlations between tests were randomlydistributed around 0.

BothGCIPLmeasurements appeared to bemore sensitiveand specific than temporal pRNFL thickness measured byCirrus or Spectralis OCT (Table 3). Estimated sensitivitiesusing Cirrus were 96.34%, 98.43%, and 58.41% for aver-age, minimum GCIPL, and temporal pRNFL, respectively.Using Spectralis, estimated sensitivity for temporal pRNFLwas lower (69.69%) than for Cirrus GCIPL measurements(average: 97.15% and minimum: 96.41%).

Importantly, average GCIPL thickness was significantlymore sensitive than temporal pRNFL by Cirrus (𝑃 < 0.05),and minimum GCIPL was significantly more sensitive thantemporal pRNFL by Spectralis for the detection of retinalthickness changes in RRMS (𝑃 < 0.05). The model appearedto be robust, as sensitivities and specificities for both GCIPLmeasurements and BCVA were similar in both models.

Abnormal results in average GCIPL thickness in RRMSpatients were independently associated to age in years (OR =0.942, 𝑃 = 0.030), years since the diagnosis of RRMS (OR =1.185, 𝑃 = 0.002), and ON antecedent (OR = 4.123, 𝑃 =0.001), after correcting by sex and intereye correlation usingbinary logistic generalized estimating equations (𝑁 = 140).

Generalised estimating equations accounting for sex, age,and within-patient intereye correlation were used to mea-sure standardised correlations between pRNFL and GCIPL


Table1:Clinicalanddemograph

iccharacteris

ticso

fhealth

ysubjectsandrelap

sing-remittingmultip

lesclerosis

(RRM

S)patie

ntsregarding

optic

neuritis(ON)antecedent.Differencesw

ere

teste

dusinggeneralized

estim

atingequatio

nsaccoun

tingforsex,age,and

with

in-patient

intereye

correla

tion.

Health

ysubjects

RRMS

ONeyes

Non

-ONeyes𝑃(health

yversus

RRMS)

𝑃(health

yversus

ONeyes)𝑃(health

yversus

non-ONeyes)

𝑃(O

Nversus

non-ONeyes)

𝑁=70

𝑁=70

𝑁=36

𝑁=104

Mean±SD

Age

(years)

37±10

40±10

42±9

40±11

0.077

0.008

0.077

0.302

Female(%)

40(57%

)44

(63%

)14

(78%

)30

(58%

)0.387

0.627

0.363

0.552

LogM

ARBC

VA0.00±0.020.04±0.180.10±0.260.02±0.14

0.00

60.023

0.115

0.083

EDSS

score

N/A

2.42±1.722.43±1.612.41±1.76

N/A

N/A

N/A

0.873

Yearssince

diagno

sisN/A

6.82±6.957.31±6.256.66±7.19

N/A

N/A

N/A

0.754

MD(in

dB)

N/A

−2.7±3.8−3.8±5.2−2.4±3.1

N/A

N/A

N/A

<0.001

Averagep

RNFL

(Sp;in𝜇m)

99.3±8.7

87.8±13.879.6±13.690.7±12.7

<0.001

<0.001

<0.001

<0.001

Tempo

ralp

RNFL

(Sp;in𝜇m)

71.4±10.6

58.8±16.050.2±16.261.9±14.8

<0.001

<0.001

<0.001

<0.001

PMBpR

NFL

(Sp;in𝜇m)

54.9±7.6

44.4±11.438.8±11.846.4±10.7

<0.001

<0.001

<0.001

<0.001

Averagep

RNFL

(Ci;in𝜇m)

93.9±8.5

84.9±12.578.9±13.087.0±11.7

<0.001

<0.001

<0.001

<0.001

Tempo

ralp

RNFL

(Ci;in𝜇m)

64.7±9.5

54.2±13.847.4±14.056.6±13.0

<0.001

<0.001

<0.001

<0.001

AverageG

CIPL

(Ci;in𝜇m)

83.8±5.9

72.2±11.366.4±10.974.3±10.7

<0.001

<0.001

<0.001

<0.001

Minim

umGCI

PL(C

i;in𝜇m)

81.9±6.1

68.1±12.961.4±13.270.4±12.0

<0.001

<0.001

<0.001

<0.001

Superio

rGCI

PL(C

i;in𝜇m)

80.5±6.2

73.0±11.766.7±11.475.2±11.1

<0.001

<0.001

<0.001

<0.001

SuperonasalG

CIPL

(Ci;in𝜇m)

85.5±6.3

73.0±11.765.8±10.875.5±11.0

<0.001

<0.001

<0.001

<0.001

InferonasalG

CIPL

(Ci;in𝜇m)

84.0±6.3

72.0±12.264.5±11.874.6±11.2

<0.001

<0.001

<0.001

<0.001

Inferio

rGCI

PL(C

i;in𝜇m)

82.7±6.2

71.6±12.065.6±12.573.7±11.1

<0.001

<0.001

<0.001

<0.001

Inferotempo

ralG

CIPL

(Ci;in𝜇m)83.8±6.1

73.3±11.068.6±12.374.9±10.1

<0.001

<0.001

<0.001

<0.001

Superotempo

ralG

CIPL

(Ci;in𝜇m)82.4±6.5

72.3±10.967.0±11.074.2±10.4

<0.001

<0.001

<0.001

<0.001

SD:stand

arddeviation;

BCVA

:bestc

orrected

visual

acuity;E

DSS:E

xpandedDisa

bilityStatus

Scale;MD:24-2SITA

Standard

Visual

FieldMeanDeviatio

n;pR

NFL

:peripapillaryretin

alnervefib

erlayer;Sp:

SpectralisOCT

;PMB:

papillo

macular

bund

le;C

i:Cirrus

OCT

;GCI

PL:ganglioncellandinnerp

lexiform

layers.


Table2:Num

bero

feyes(andpercentage)w

ithabno

rmalresults

(yellowor

redcolor-coded)

inhealthyandrelap

sing-remittingmultip

lesclerosis

(RRM

S)patie

ntsa

ndaccordingto

optic

neuritis(ON)a

ntecedent.Statisticalsig

nificance

ofassociations

was

teste

dusinggeneralized

estim

atingequatio

nsaccoun

tingforsex,age,and

with

in-patient

intereye

correlation.

Abno

rmalresult

Health

yRR

MS

ONeyes

Non

-ONeyes𝑃(health

yversus

RRMS)

𝑃(health

yversus

ONeyes)𝑃(health

yversus

non-ONeyes)

𝑃(O

Nversus

non-ONeyes)

𝑁=140𝑁=140𝑁=36

𝑁=104

Averagep

RNFL

(Sp)

5(3.6)

49(35.0)

20(55.6)

29(27.9

)<0.001

<0.001

<0.001

<0.001

Tempo

ralp

RNFL

(Sp)

5(3.6)

65(46.4)

22(61.1)

43(41.3

)<0.001

<0.001

<0.001

<0.001

PMBpR

NFL

(Sp)

4(2.9)

43(30.7)

17(47.2

2)26

(25.0)

<0.001

<0.001

<0.001

0.012

Averagep

RNFL

(Ci)

9(6.4)

46(32.9)

19(52.8)

27(26.0)

<0.001

<0.001

0.001

0.001

Tempo

ralp

RNFL

(Ci)

2(1.4)

49(35.0)

24(66.7)

25(24.0)

<0.001

<0.001

<0.001

<0.001

AverageG

CIPL

(Ci)

8(5.7)

65(46.4)

25(69.4

)40

(38.5)

<0.001

<0.001

<0.001

<0.001

Minim

umGCI

PL(C

i)7(5.0)

67(47.9

)28

(77.8

)39

(37.5

)<0.001

<0.001

<0.001

<0.001

Superio

rGCI

PL(C

i)8(5.7)

63(45.0)

26(72.2)

37(35.6)

<0.001

<0.001

<0.001

<0.001

SuperonasalG

CIPL

(Ci)

4(2.9)

66(47.1)

26(72.2)

40(38.5)

<0.001

<0.001

<0.001

<0.001

InferonasalG

CIPL

(Ci)

4(2.9)

58(41.4

)27

(75.0)

31(29.8

)<0.001

<0.001

<0.001

<0.001

Inferio

rGCI

PL(C

i)2(1.4)

50(35.7)

23(63.9)

27(26.0)

<0.001

<0.001

<0.001

<0.001

Inferotempo

ralG

CIPL

(Ci)

6(4.3)

55(39.3

)22

(61.1)

33(31.7

)<0.001

<0.001

<0.001

0.002

Superotempo

ralG

CIPL

(Ci)

12(8.9)

64(45.7)

25(69.4

)39

(37.5

)<0.001

<0.001

<0.001

<0.001

pRNFL

:peripapillaryretin

alnervefi

berlayer;Sp:SpectralisOCT

;PMB:

papillo

macular

bund

le;C

i:Cirrus

OCT

;GCI

PL:ganglioncellandinnerp

lexiform

layers.


0102030405060708090

100

Aver

age

Min

imum

Supe

rior

Supe

rona

sal

Infe

rona

sal

Infe

rior

Infe

rote

mpo

ral

Supe

rote

mpo

ral

Aver

age

Min

imum

Supe

rior

Supe

rona

sal

Infe

rona

sal

Infe

rior

Infe

rote

mpo

ral

Supe

rote

mpo

ral

Aver

age

Min

imum

Supe

rior

Supe

rona

sal

Infe

rona

sal

Infe

rior

Infe

rote

mpo

ral

Supe

rote

mpo

ral

Eyes without optic neuritis Eyes with optic neuritis Control eyes

WhiteGreen

YellowRed

(%)

Figure 2: Comparison of the color scale frequency for each sector using ganglion cell-inner plexiform layer analysis with Cirrus opticalcoherence tomography among eyes with and without optic neuritis in relapsing-remittingmultiple sclerosis patients and from healthy controlsubjects. Black represents eyes classified as red (below percentile 1); dark gray represents eyes labeled as yellow (below percentile 5); light grayrepresents green (between percentiles 5 and 95), and white represents white (above percentile 95).

thicknesses and disease duration, EDSS, and visual functionparameters (Table 4).

Average pRNFL thickness measured with Cirrus (𝛽 =−0.233; 𝑃 = 0.031) and Spectralis (𝛽 = −0.228; 𝑃 = 0.025)OCTs, temporal (𝛽 = −0.261; 𝑃 = 0.017) and papillomacularbundle (𝛽 = −0.275; 𝑃 = 0.006) pRNFL thickness measuredwith Spectralis OCT, average GCIPL thickness (𝛽 = −0.262;𝑃 = 0.026), and minimum GCIPL thickness (𝛽 = −0.299;𝑃 = 0.008) correlated inversely with disease duration.

A strong, positive correlation was observed betweenaverage GCIPL thickness and pRNFL thickness using Cirrus(𝛽 = 0.692; 𝑃 < 0.001) and Spectralis (𝛽 = 0.642; 𝑃 < 0.001)OCTs.

After adjusting by age, sex, and within-patient intereyecorrelation, Cirrus GCIPL average and minimum measuresshowed a weak but significant correlation with EDSS only ineyes with ON antecedent (Table 4).

4. Discussion

Loss of RNFL is a well-documented structural marker ofaxonal degeneration in the eyes of patients with multiplesclerosis with and without a history of ON [8, 24–26].Historically, OCT studies in multiple sclerosis have focusedmostly on the RNFL, but retinal ganglion cell neuronalloss may also be implicated in the pathogenesis of visualdysfunction in MS [10].

To the best of our knowledge, apart from direct compar-isons between pRNFL and GCIPL measurements in RRMSand healthy eyes [7–10], there is no information about thesensitivity and specificity of the different measurements todetect retinal thickness changes in RRMS patients.

Although strong positive correlation was found betweenGCIPL and pRNFL thickness values, average and mini-mum GCIPL measures showed higher sensitivity and speci-ficity than temporal pRNFL measures. Remarkably, average

GCIPL showed a significantly better sensitivity than temporalpRNFL by Cirrus, and minimum GCIPL showed a signifi-cantly better sensitivity than temporal pRNFL by Spectralisfor the detection of retinal thickness changes in RRMS.

In agreement with previous studies [21], we found thateyes from patients with RRMS had significant thinning inaverage and temporal quadrant pRNFL values by Cirrusand Spectralis OCT, and, in each of six sectors, averageand minimum GCIPL results obtained by Cirrus, whencompared to healthy eyes (Table 1). This was true for eyeswith andwithout ON antecedent. Nevertheless, eyes withONhistory showed greater pRNFL andGCIPL thinning than eyeswithout ON antecedent [10] (Table 1).

In a previous study [4] analyzing the color-code clas-sification of pRNFL in RRMS patients, we identified thetemporal quadrant to be the most abnormally color-codedby both Cirrus and Spectralis. Additionally, temporal pRNFLquadrants were abnormally color-coded more frequently inON eyes than in non-ON eyes by both devices, Cirrus andSpectralis.

The abnormality rate in temporal pRNFL color codein eyes with previous history of ON was 66.7% by Cirrusand 61.1% by Spectralis. The current study agrees withprevious reports that eyes from patients with RRMS exhibita significant thinning of the pRNFL and GCIPL comparedwith the healthy eyes [7, 10, 11] and that pRNFL thinning inRRMS patients typically occurs in the temporal sector [4, 20–23, 27–29].

As expected, the sector showing the highest abnormalityrate in GCIPL test was the superonasal (47.1%).

Unsurprisingly, GCIPL thinning showed an associationwith disease duration and ON antecedent in eyes of patientswith RRMS. GCIPL minimum represents the lowest GCIPLthickness over a single meridian crossing the annulus, whichis expected to be sensitive to focal damage [30].Thismeasure-ment had previously been found to have the highest correla-tion with visual field pattern standard deviation in patients


Table 3: Estimated sensitivity and specificity of different optic coherence tomography measures for retinal thickness changes detection ineyes of patients with relapsing-remitting multiple sclerosis (𝑁 = 140) and from healthy control subjects (𝑁 = 140) using latent class analysis.Model A: temporal pRNFL thickness asmeasured byCirrusOCT; estimated prevalence of retinal thickness changes in the sample was 23.42%;95% confidence interval (95CI) 18.50 to 29.17%. Model B: temporal pRNFL thickness as measured by Spectralis OCT; estimated prevalenceof retinal thickness changes in the sample was 24.25%; 95% confidence interval (95CI) 19.26 to 30.04%.

Sensitivity SpecificityEstimate 95CI Estimate 95CI

Model ACirrus average GCIPL 96.34% 76.11 to 99.54% 97.05% 92.77 to 98.83%Cirrus minimum GCIPL 98.43% 64.61 to 99.95% 95.81% 91.00 to 98.10%Cirrus temporal pRNFL 58.41% 45.82 to 70.00% 93.91% 89.56 to 96.52%Best corrected visual acuity 6.14% 2.32 to 15.30% 98.59% 95.72 to 99.54%

Model BCirrus average GCIPL 97.15% 76.94 to 99.71% 97.61% 93.16 to 99.19%Cirrus minimum GCIPL 96.41% 81.48 to 99.39% 95.43% 90.80 to 97.78%Spectralis temporal pRNFL 69.69% 57.14 to 79.86% 92.70% 88.01 to 95.65%Best corrected visual acuity 6.08% 2.30 to 15.15% 98.55% 95.58 to 99.53%

GCIPL: ganglion cell-inner plexiform layers; pRNFL: peripapillary retinal nerve fiber layer.

Table 4: Standardized correlation coefficients betweenOCTmeasurements and neurologic and visual function parameters in eyes of patientswith relapsing-remitting multiple sclerosis, calculated using generalized estimating equations accounting for sex, age, and within-patientintereye correlation (𝑛 = 140).

BCVA MD EDSS Disease durationCirrus average pRNFL 0.286∗ 0.418∗ −0.014 −0.233‡

Cirrus temporal pRNFL −0.013 0.062 0.001 −0.141Spectralis average pRNFL 0.314∗ 0.360∗ 0.033 −0.228‡

Spectralis temporal pRNFL 0.122‡ 0.268† −0.101 −0.261‡

Spectralis PMB pRNFL 0.125‡ 0.209‡ −0.066 −0.275†

GCIPL average 0.226∗ 0.513∗ −0.178 −0.262‡

GCIPL minimum 0.204† 0.412∗ −0.161 −0.299†

Non-ON eyes (𝑁 = 104)Cirrus average pRNFL −0.019 0.098 0.033 −0.253‡

Cirrus temporal pRNFL −0.270‡ −0.030 −0.046 −0.146Spectralis average pRNFL 0.210∗ 0.145 0.109 −0.274‡

Spectralis temporal pRNFL −0.025 0.093 −0.092 −0.284†

Spectralis PMB pRNFL 0.002 0.083 −0.040 −0.332†

GCIPL average −0.062 0.109 −0.113 −0.297‡

GCIPL minimum 0.029 0.138 −0.092 −0.314†

ON eyes (𝑁 = 36)Cirrus average pRNFL 0.188 0.161 −0.173 0.012Cirrus temporal pRNFL −0.043 0.001 0.105 −0.013Spectralis average pRNFL 0.175 0.149 −0.151 0.109Spectralis temporal pRNFL 0.112 0.030 −0.195 0.065Spectralis PMB pRNFL 0.059 −0.022 −0.120 0.097GCIPL average 0.145 0.133 −0.429† −0.079GCIPL minimum 0.144 0.120 −0.421‡ −0.227∗

𝑃 < 0.001; †𝑃 < 0.01; ‡𝑃 < 0.05.OCT: optic coherence tomography; BCVA: best corrected visual acuity; MD: Goldman 24-2 SITA standard visual field mean deviation; EDSS: ExpandedDisability Status Scale; pRNFL: peripapillary retinal nerve fiber layer; PMB: papillomacular bundle; GCIPL: ganglion cell and inner plexiform layers; ON: opticneuritis.


with chronic open angle glaucoma [31]. In our study, CirrusGCIPL minimum measures showed a significant correlationwith BCVA, mean deviation, and EDSS. Importantly, onlyGCIPL measures showed significant correlation with EDSSin ON eyes (Table 4).

A superior structure-function correlation betweenGCIPL thickness and clinical measures compared to pRNFLhas been reported recently, suggesting that GCIPL analysismight be a better approach than pRNFL to examine MSneurodegeneration [11].

Previous studies have shown that GCIPL thickness canbe altered in patients with RRMS [10, 32–34] and that thesealterations correlated with visual function [10, 11] and centralnervous system findings using magnetic resonance imaging[28]. Some of these studies have suggested that GCIPLthickness can be a more reliable measure for the detectionof retinal anomalies in RRMS than pRNFL thickness [10].Interestingly, our study demonstrates that a decrease inGCIPL thickness is more sensitive to retinal involvement inRRMS than an alteration in the temporal pRNFL.

This study has a number of limitations warranting dis-cussion. Firstly, exact sensitivity and specificity cannot beobtained without a gold standard test. The values obtainedthrough LCA can be useful when comparing different tests;however, the sensitivities and specificities provided are esti-mates. Studies with bigger sample size and including infor-mation on other diagnostic tests, such as electrodiagnostictesting and magnetic resonance imaging, could help toimprove this estimation. Secondly, both eyes were included inthis study. However, most MS studies published to date haveincluded both eyes because they can be individually evaluatedand do not necessarily follow the same disease course [1, 4].Additionally, generalized estimating equations were used inorder to account for sex, age, and within-patient intereyecorrelation.With these methods, information from both eyescan be used for the study, without the risk of increasing therisk of bias due to intereye correlation or increase in samplesize [35].

The retinal segmentation algorithm used by Cirrus OCTcombines GCL and IPL, since the boundaries between thesetwo layers cannot be visually discriminated on this device.Although Spectralis OCT has developed a specific softwarethat provides automated differentiation and quantification ofthe individual retinal layers, it was not available when the datawas collected. Additionally, this software does not providecomparison to a normative database, so binary outcomesnecessary for LCA would not be available.

Finally, we have included only patients with relapsing-remitting multiple sclerosis; therefore, our results cannot beextrapolated to other types of multiple sclerosis or to patientswith more advanced disease (mean EDSS was 2.42).

5. Conclusions

In conclusion, OCT GCIPL analysis is more sensitive thantemporal pRNFL analysis to detect retinal thickness changesin RRMS eyes. GCIPL measures correlate better than pRNFLmeasures with visual function parameters such as BCVA,visual field mean deviation, or EDSS. As such, GCIPL

thickness measured by retinal segmentation of OCT scansmay be an ideal marker for monitoring neurodegenerationin RRMS patients.

Conflict of Interests

Theauthors have noproprietary or commercial interest in anymaterials discussed in this paper. All authors have completedand submitted the ICMJE form for disclosure of potentialconflict of interests and none were reported.

Authors’ Contribution

Conception and design was done by Julio J. Gonzalez-Lopez, GemaRebolleda, Francisco J.Munoz-Negrete,MarinaLeal, Noelia Oblanca, Lucienne Costa-Frossard, and Jose C.Alvarez-Cermeno; analysis and interpretation was done byJulio J. Gonzalez-Lopez and Gema Rebolleda; writing thepaper was done by Julio J. Gonzalez-Lopez, Gema Rebolleda,and Francisco J. Munoz-Negrete; critical revision of thepaper was done by Marina Leal, Noelia Oblanca, LucienneCosta-Frossard, and Jose C. Alvarez-Cermeno. Final approvalof the paper was done by Julio J. Gonzalez-Lopez, GemaRebolleda, Francisco J. Munoz-Negrete, Marina Leal, NoeliaOblanca, Lucienne Costa-Frossard, and Jose C. Alvarez-Cermeno; data collection was done by Julio J. Gonzalez-Lopez, Marina Leal, and Noelia Oblanca; provision of mate-rials, patients, or resources was done by Gema Rebolleda,Francisco J. Munoz-Negrete, Lucienne Costa-Frossard, andJose C. Alvarez-Cermeno; statistical expertise was providedby Julio J. Gonzalez-Lopez; obtaining funding was doneby Gema Rebolleda and Francisco J. Munoz-Negrete; andliterature search was done by Julio J. Gonzalez-Lopez andGema Rebolleda.

Acknowledgments

Julio J. Gonzalez-Lopez is a Ph.D. candidate at the SurgeryDepartment, Universidad de Alcala School of Medicine,Madrid, Spain. This study has been granted by the “Fondode Investigaciones Sanitarias,” from the Spanish Ministry ofHealth and Consumer Affairs.

References

[1] E. Ojeda, D. Dıaz-Cortes, D. Rosales, C. Duarte-Rey, J.-M.Anaya, and A. Rojas-Villarraga, “Prevalence and clinical fea-tures of multiple sclerosis in Latin America,” Clinical Neurologyand Neurosurgery, vol. 115, no. 4, pp. 381–387, 2013.

[2] F. Ikuta and H. M. Zimmerman, “Distribution of plaques inseventy autopsy cases of multiple sclerosis in the United States,”Neurology, vol. 26, part 2, no. 6, pp. 26–28, 1976.

[3] R. H. Kardon, “Role of the macular optical coherencetomography scan in neuro-ophthalmology,” Journal of Neuro-Ophthalmology, vol. 31, no. 4, pp. 353–361, 2011.

[4] G. Rebolleda, J. J. Gonzalez-Lopez, F. J. Munoz-Negrete, N.Oblanca, L. Costa-Frossard, and J. C. Alvarez-Cermeno, “Color-code agreement among stratus, cirrus, and spectralis opticalcoherence tomography in relapsing-remittingmultiple sclerosis


with and without prior optic neuritis,”The American Journal ofOphthalmology, vol. 155, no. 5, pp. 890–897, 2013.

[5] A. J. Green, S. McQuaid, S. L. Hauser, I. V. Allen, and R. Lyness,“Ocular pathology in multiple sclerosis: retinal atrophy andinflammation irrespective of disease duration,” Brain, vol. 133,part 6, pp. 1591–1601, 2010.

[6] K. S. Shindler, E. Ventura,M. Dutt, and A. Rostami, “Inflamma-tory demyelination induces axonal injury and retinal ganglioncell apoptosis in experimental optic neuritis,” Experimental EyeResearch, vol. 87, no. 3, pp. 208–213, 2008.

[7] P. Albrecht, M. Ringelstein, A. K.Muller et al., “Degeneration ofretinal layers inmultiple sclerosis subtypes quantified by opticalcoherence tomography,” Multiple Sclerosis, vol. 18, no. 10, pp.1422–1429, 2012.

[8] S. B. Syc, S. Saidha, S. D. Newsome et al., “Optical coherencetomography segmentation reveals ganglion cell layer pathologyafter optic neuritis,” Brain, vol. 135, part 2, pp. 521–533, 2012.

[9] L. S. Talman, E. R. Bisker, D. J. Sackel et al., “Longitudinalstudy of vision and retinal nerve fiber layer thickness inmultiplesclerosis,” Annals of Neurology, vol. 67, no. 6, pp. 749–760, 2010.

[10] S. D. Walter, H. Ishikawa, K. M. Galetta et al., “Ganglioncell loss in relation to visual disability in multiple sclerosis,”Ophthalmology, vol. 119, no. 6, pp. 1250–1257, 2012.

[11] S. Saidha, S. B. Syc, M. K. Durbin et al., “Visual dysfunctionin multiple sclerosis correlates better with optical coherencetomography derived estimates of macular ganglion cell layerthickness than peripapillary retinal nerve fiber layer thickness,”Multiple Sclerosis, vol. 17, no. 12, pp. 1449–1463, 2011.

[12] C. H. Polman, S. C. Reingold, G. Edan et al., “Diagnostic criteriaformultiple sclerosis: 2005 revisions to the ‘McDonaldCriteria’,”Annals of Neurology, vol. 58, no. 6, pp. 840–846, 2005.

[13] M. Sadatsafavi, N. Shahidi, F. Marra et al., “A statistical methodwas used for the meta-analysis of tests for latent TB in theabsence of a gold standard, combining random-effect andlatent-class methods to estimate test accuracy,” Journal ofClinical Epidemiology, vol. 63, no. 3, pp. 257–269, 2010.

[14] S. L. Hui and S. D. Walter, “Estimating the error rates ofdiagnostic tests,” Biometrics, vol. 36, no. 1, pp. 167–171, 1980.

[15] D. Rindskopf and W. Rindskopf, “The value of latent classanalysis in medical diagnosis,” Statistics in Medicine, vol. 5, no.1, pp. 21–27, 1986.

[16] E. Girardi, C. Angeletti, V. Puro et al., “Estimating diagnosticaccuracy of tests for latent tuberculosis infection without a goldstandard among healthcare workers,” Euro Surveillance, vol. 14,no. 43, 2009.

[17] T. S. Machado de Assis, A. Rabello, and G. L. Werneck, “Latentclass analysis of diagnostic tests for visceral leishmaniasis inBrazil,” Tropical Medicine and International Health, vol. 17, no.10, pp. 1202–1207, 2012.

[18] M. Shevlin, S. Murphy, and J. Murphy, “Adolescent lonelinessand psychiatricmorbidity in the general population: Identifying“at risk” groups using latent class analysis,” Nordic Journal ofPsychiatry, 2014.

[19] M. Ang, W. L. Wong, X. Li, and S.-P. Chee, “Interferon 𝛾 releaseassay for the diagnosis of uveitis associated with tuberculosis:a Bayesian evaluation in the absence of a gold standard,” TheBritish Journal of Ophthalmology, vol. 97, no. 8, pp. 1062–1067,2013.

[20] C. Fjeldstad, M. Bemben, and G. Pardo, “Reduced retinal nervefiber layer and macular thickness in patients with multiplesclerosis with no history of optic neuritis identified by the use

of spectral domain high-definition optical coherence tomogra-phy,” Journal of Clinical Neuroscience, vol. 18, no. 11, pp. 1469–1472, 2011.

[21] E. Garcia-Martin, V. Pueyo, I. Pinilla, J.-R. Ara, J. Martin,and J. Fernandez, “Fourier-domain OCT in multiple sclerosispatients: reproducibility and ability to detect retinal nerve fiberlayer atrophy,” Investigative Ophthalmology and Visual Science,vol. 52, no. 7, pp. 4124–4131, 2011.

[22] S. Noval, I. Contreras, S. Munoz, C. Oreja-Guevara, B. Man-zano, and G. Rebolleda, “Optical coherence tomography inmultiple sclerosis and neuromyelitis optica: an update,”MultipleSclerosis International, vol. 2011, Article ID 472790, 11 pages,2011.

[23] G. Rebolleda, A. Garcıa-Garcıa, H. R.WonKim, and F. J. M��oz-Negrete, “Comparison of retinal nerve fiber layer measured bytime domain and spectral domain optical coherence tomogra-phy in optic neuritis,” Eye, vol. 25, no. 2, pp. 233–238, 2011.

[24] K. M. Galetta, P. A. Calabresi, E. M. Frohman, and L. J. Balcer,“Optical coherence tomography (OCT): imaging the visualpathway as a model for neurodegeneration,” Neurotherapeutics,vol. 8, no. 1, pp. 117–132, 2011.

[25] A. Petzold, J. F. de Boer, S. Schippling et al., “Optical coherencetomography inmultiple sclerosis: a systematic review andmeta-analysis,”The Lancet Neurology, vol. 9, no. 9, pp. 921–932, 2010.

[26] S. Saidha, S. B. Syc, M. A. Ibrahim et al., “Primary retinalpathology in multiple sclerosis as detected by optical coherencetomography,” Brain, vol. 134, part 2, pp. 518–533, 2011.

[27] J. B. Fisher, D. A. Jacobs, C. E. Markowitz et al., “Relation ofvisual function to retinal nerve fiber layer thickness in multiplesclerosis,” Ophthalmology, vol. 113, no. 2, pp. 324–332, 2006.

[28] E. Garcıa-Martın, V. Pueyo, J. Martin et al., “Progressivechanges in the retinal nerve fiber layer in patients with multiplesclerosis,” European Journal of Ophthalmology, vol. 20, no. 1, pp.167–173, 2010.

[29] V. Pueyo, J. R. Ara, C. Almarcegui et al., “Sub-clinical atrophyof the retinal nerve fibre layer in multiple sclerosis,” ActaOphthalmologica, vol. 88, no. 7, pp. 748–752, 2010.

[30] J.-C. Mwanza, M. K. Durbin, D. L. Budenz et al., “Profileand predictors of normal ganglion cell-inner plexiform layerthickness measured with frequency-domain optical coherencetomography,” Investigative Ophthalmology &Visual Science, vol.52, no. 11, pp. 7872–7879, 2011.

[31] H. El Chehab, M. Delbarre, J. Fenolland et al., “Structurefunction relationship: specific index of ganglion cell complexassessment of 2 spectral domain OCT and standard automatedperimetry,” Acta Ophthalmologica Scandinavica, vol. 91, supple-ment s252, 2013.

[32] D. B. Fernandes, A. S. Raza, R. G. Nogueira et al., “Evaluationof inner retinal layers in patients with multiple sclerosis orneuromyelitis optica using optical coherence tomography,”Ophthalmology, vol. 120, no. 2, pp. 387–394, 2013.

[33] J. N. Ratchford, S. Saidha, E. S. Sotirchos et al., “Active MSis associated with accelerated retinal ganglion cell/inner plex-iform layer thinning,” Neurology, vol. 80, no. 1, pp. 47–54, 2013.

[34] S. Saidha, E. S. Sotirchos, J. Oh et al., “Relationships betweenretinal axonal and neuronal measures and global centralnervous system pathology in multiple sclerosis,” Archives ofNeurology, vol. 70, no. 1, pp. 34–43, 2013.

[35] I. E. Murdoch, S. S. Morris, and S. N. Cousens, “People andeyes: statistical approaches in ophthalmology,” British Journalof Ophthalmology, vol. 82, no. 8, pp. 971–973, 1998.

Submit your manuscripts athttp://www.hindawi.com

Stem CellsInternational

Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014


MEDIATORSINFLAMMATION

of


Behavioural Neurology

EndocrinologyInternational Journal of



Disease Markers


BioMed Research International

OncologyJournal of



Oxidative Medicine and Cellular Longevity


PPAR Research

The Scientific World JournalHindawi Publishing Corporation http://www.hindawi.com Volume 2014

Immunology ResearchHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014

Journal of

ObesityJournal of



Computational and Mathematical Methods in Medicine

OphthalmologyJournal of


Diabetes ResearchJournal of



Research and TreatmentAIDS


Gastroenterology Research and Practice


Parkinson’s Disease

Evidence-Based Complementary and Alternative Medicine

Volume 2014Hindawi Publishing Corporationhttp://www.hindawi.com

clinical study comparative diagnostic accuracy of ganglion ...clinical study comparative diagnostic...

Documents